Generating synthetic computed tomography for radiotherapy: SynthRAD2023 challenge report.

Radiation therapy plays a crucial role in cancer treatment, necessitating precise delivery of radiation to tumors while sparing healthy tissues over multiple days. Computed tomography (CT) is integral for treatment planning, offering electron density data crucial for accurate dose calculations. However, accurately representing patient anatomy is challenging, especially in adaptive radiotherapy, where CT is not acquired daily. Magnetic resonance imaging (MRI) provides superior soft-tissue contrast. Still, it lacks electron density information, while cone beam CT (CBCT) lacks direct electron density calibration and is mainly used for patient positioning. Adopting MRI-only or CBCT-based adaptive radiotherapy eliminates the need for CT planning but presents challenges. Synthetic CT (sCT) generation techniques aim to address these challenges by using image synthesis to bridge the gap between MRI, CBCT, and CT. The SynthRAD2023 challenge was organized to compare synthetic CT generation methods using multi-center ground truth data from 1080 patients, divided into two tasks: (1) MRI-to-CT and (2) CBCT-to-CT. The evaluation included image similarity and dose-based metrics from proton and photon plans. The challenge attracted significant participation, with 617 registrations and 22/17 valid submissions for tasks 1/2. Top-performing teams achieved high structural similarity indices (≥0.87/0.90) and gamma pass rates for photon (≥98.1%/99.0%) and proton (≥97.3%/97.0%) plans. However, no significant correlation was found between image similarity metrics and dose accuracy, emphasizing the need for dose evaluation when assessing the clinical applicability of sCT. SynthRAD2023 facilitated the investigation and benchmarking of sCT generation techniques, providing insights for developing MRI-only and CBCT-based adaptive radiotherapy. It showcased the growing capacity of deep learning to produce high-quality sCT, reducing reliance on conventional CT for treatment planning.

Relating neural oscillations to laminar fMRI connectivity in visual cortex.

Laminar functional magnetic resonance imaging (fMRI) holds the potential to study connectivity at the laminar level in humans. Here we analyze simultaneously recorded electroencephalography (EEG) and high-resolution fMRI data to investigate how EEG power modulations, induced by a task with an attentional component, relate to changes in fMRI laminar connectivity between and within brain regions in visual cortex. Our results indicate that our task-induced decrease in beta power relates to an increase in deep-to-deep layer coupling between regions and to an increase in deep/middle-to-superficial layer connectivity within brain regions. The attention-related alpha power decrease predominantly relates to reduced connectivity between deep and superficial layers within brain regions, since, unlike beta power, alpha power was found to be positively correlated to connectivity. We observed no strong relation between laminar connectivity and gamma band oscillations. These results indicate that especially beta band, and to a lesser extent, alpha band oscillations relate to laminar-specific fMRI connectivity. The differential effects for alpha and beta bands indicate that they relate to different feedback-related neural processes that are differentially expressed in intra-region laminar fMRI-based connectivity.

Magnetization transfer weighted laminar fMRI with multi-echo FLASH.

Laminar functional magnetic resonance imaging (fMRI) using the gradient echo (GRE) blood oxygenation level dependent (BOLD) contrast is prone to signal changes arising from large unspecific venous vessels. Alternatives based on changes of cerebral blood volume (CBV) become more popular since it is expected that this hemodynamic response is dominant in microvasculature. One approach to sensitize the signal toward changes in CBV, and to simultaneously reduce unwanted extravascular (EV) BOLD blurring, is to selectively reduce gray matter (GM) signal via magnetization transfer (MT). In this work, we use off-resonant MT-pulses with a 3D FLASH readout to perform MT-prepared (MT-prep) laminar fMRI of the primary visual cortex (V1) at multiple echo times at 7 T. With a GRE-BOLD contrast without additional MT-weighting as reference, we investigated the influence of the MT-preparation on the shape and the echo time dependency of laminar profiles. Through numerical simulations, we optimized the sequence parameters to increase the sensitivity toward signal changes induced by changes in arterial CBV and to delineate the contributions of different compartments to the signal. We show that at 7 T, GM signals can be reduced by 30 %. Our laminar fMRI responses exhibit an increased signal change in the parenchyma at very short TE compared to a BOLD-only reference as a result of reduced EV signal intensity. By varying echo times, we could show that MT-prep results in less sensitivity toward unwanted signal changes based on changes in T2*. We conclude that when accounting for nuclear overhauser enhancement effects in blood, off-resonant MT-prep combined with efficient short TE readouts can become a promising method to reduce unwanted EV venous contributions in GRE-BOLD and/or to allow scanning at much shorter echo times without incurring a sensitivity penalty in laminar fMRI.

Laminar fMRI using T2-prepared multi-echo FLASH.

Functional magnetic resonance imaging (fMRI) using blood oxygenation level dependent (BOLD) contrast at a sub-millimeter scale is a promising technique to probe neural activity at the level of cortical layers. While gradient echo (GRE) BOLD sequences exhibit the highest sensitivity, their signal is confounded by unspecific extravascular (EV) and intravascular (IV) effects of large intracortical ascending veins and pial veins leading to a downstream blurring effect of local signal changes. In contrast, spin echo (SE) fMRI promises higher specificity towards signal changes near the microvascular compartment. However, the T2-weighted signal is typically sampled with a gradient echo readout imposing additional T2'-weighting. In this work, we used a T2-prepared (T2-prep) sequence with short GRE readouts to investigate its capability to acquire laminar fMRI data during a visual task in humans at 7 T. By varying the T2-prep echo time (TEprep) and acquiring multiple gradient echoes (TEGRE) per excitation, we studied the specificity of the sequence and the influence of possible confounding contributions to the shape of laminar fMRI profiles. By fitting and extrapolating the multi-echo GRE data to a TEGRE = 0 ms condition, we show for the first time laminar profiles free of T2'-pollution, confined to gray matter. This finding is independent of TEprep, except for the shortest one (31 ms) where hints of a remaining intravascular component can be seen. For TEGRE > 0 ms a prominent peak at the pial surface is observed that increases with longer TEGRE and dominates the shape of the profiles independent of the amount of T2-weighting. Simulations show that the peak at the pial surface is a result of static EV dephasing around pial vessels in CSF visible in GM due to partial voluming. Additionally, another, weaker, static dephasing effect is observed throughout all layers of the cortex, which is particularly obvious in the data with shortest T2-prep echo time. Our simulations show that this cannot be explained by intravascular dephasing but that it is likely caused by extravascular effects of the intracortical and pial veins. We conclude that even for TEGRE as short as 2.3 ms, the T2'-weighting added to the T2-weighting is enough to dramatically affect the laminar specificity of the BOLD signal change. However, the bulk of this corruption stems from CSF partial volume effects which can in principle be addressed by increasing the spatial resolution of the acquisition.

Enhanced POCS reconstruction for partial Fourier imaging in multi-echo and time-series acquisitions.

PURPOSE: To improve partial Fourier (PF) imaging reconstruction in time-series or multi-echo acquisitions. METHODS: Many PF methods use a phase estimate to restore Hermitian symmetry before filling missing k-space entries with measured data from the opposite half. This estimate is obtained from the symmetrically sampled, central part of k-space and its low-resolution results in artifacts near high-frequency phase effects (eg, tissue boundaries, vessels), limiting PF undersampling. Enhanced projection onto convex sets (POCS) uses full-resolution phase estimates and relies on alternating the half of k-space that is acquired in time series or multi-echo acquisitions. This enables full-resolution phase estimates to be calculated for each volume/echo, which are fed into the POCS framework. We apply enhanced POCS to high-resolution multi-echo FLASH and 3D-EPI functional MRI time-series data. RESULTS: Reconstruction errors and their bias dramatically reduce compared with existing methods, without leading to temporal blurring in time-series acquisitions. This allows for higher PF acceleration factors at virtually no cost. CONCLUSION: Enhanced POCS results in superior PF reconstructions. Furthermore, as the resolution of the phase estimate used for symmetry correction no longer depends on the PF factor, enhanced POCS is more robust against larger PF omission.

Improved cortical boundary registration for locally distorted fMRI scans.

With continuing advances in MRI techniques and the emergence of higher static field strengths, submillimetre spatial resolution is now possible in human functional imaging experiments. This has opened up the way for more specific types of analysis, for example investigation of the cortical layers of the brain. With this increased specificity, it is important to correct for the geometrical distortions that are inherent to echo planar imaging (EPI). Inconveniently, higher field strength also increases these distortions. The resulting displacements can easily amount to several millimetres and as such pose a serious problem for laminar analysis. We here present a method, Recursive Boundary Registration (RBR), that corrects distortions between an anatomical and an EPI volume. By recursively applying Boundary Based Registration (BBR) on progressively smaller subregions of the brain we generate an accurate whole-brain registration, based on the grey-white matter contrast. Explicit care is taken that the deformation does not break the topology of the cortical surface, which is an important requirement for several of the most common subsequent steps in laminar analysis. We show that RBR obtains submillimetre accuracy with respect to a manually distorted gold standard, and apply it to a set of human in vivo scans to show a clear increase in spacial specificity. RBR further automates the process of non-linear distortion correction. This is an important step towards routine human laminar fMRI for large field of view acquisitions. We provide the code for the RBR algorithm, as well as a variety of functions to better investigate registration performance in a public GitHub repository, https://github.com/TimVanMourik/OpenFmriAnalysis, under the GPL 3.0 license.

Diffusion Acceleration with Gaussian process Estimated Reconstruction (DAGER).

PURPOSE: Image acceleration provides multiple benefits to diffusion MRI, with in-plane acceleration reducing distortion and slice-wise acceleration increasing the number of directions that can be acquired in a given scan time. However, as acceleration factors increase, the reconstruction problem becomes ill-conditioned, particularly when using both in-plane acceleration and simultaneous multislice imaging. In this work, we develop a novel reconstruction method for in vivo MRI acquisition that provides acceleration beyond what conventional techniques can achieve. THEORY AND METHODS: We propose to constrain the reconstruction in the spatial (k) domain by incorporating information from the angular (q) domain. This approach exploits smoothness of the signal in q-space using Gaussian processes, as has previously been exploited in post-reconstruction analysis. We demonstrate in-plane undersampling exceeding the theoretical parallel imaging limits, and simultaneous multislice combined with in-plane undersampling at a total factor of 12. This reconstruction is cast within a Bayesian framework that incorporates estimation of smoothness hyper-parameters, with no need for manual tuning. RESULTS: Simulations and in vivo results demonstrate superior performance of the proposed method compared with conventional parallel imaging methods. These improvements are achieved without loss of spatial or angular resolution and require only a minor modification to standard pulse sequences. CONCLUSION: The proposed method provides improvements over existing methods for diffusion acceleration, particularly for high simultaneous multislice acceleration with in-plane undersampling.

Strategies and prospects for cortical depth dependent T2 and T2* weighted BOLD fMRI studies.

Technological advancements in fMRI have afforded the opportunity to conduct submillimeter investigations into human brain function. The ability to do cortical depth dependent (or layer-specific) fMRI could allow probing intrinsic neuronal organizations and inter-connections, including the directionality of interregional information flow, while ultimately shedding light on uniquely human behaviors. The methodological development and applications of cortical depth dependent fMRI has been ongoing for nearly a decade, yet a consensus on protocols, analysis pipelines and interpretations of data has yet to be reached. In this article we discuss the current status and understandings of high resolution fMRI data and methods. In particular, the spatial signal characteristics of the BOLD contrast are examined in relation to the intra-cortical vasculature and associated underlying contrast mechanisms for three pulse sequences used in laminar fMRI: gradient echo (GE), spin echo (SE) and 3D-GRASE. We conclude with a short overview of factors to be taken into account when designing a specific study, including the choice of pulse sequence, experimental design, and data analysis strategy.

Laminar Organization of Working Memory Signals in Human Visual Cortex.

The human primary visual cortex (V1) is not only activated by incoming visual information but is also engaged by top-down cognitive processes, such as visual working memory, even in the absence of visual input [1-3]. This feedback may be critical to our ability to visualize specific visual features, as higher-order regions lack the selectivity to represent such information [4]. Clearly, such internally generated signals do not trigger genuine perception of the remembered stimulus, meaning they must be organized in a manner that is different to bottom-up-driven signals. Internally generated signals may be kept separate from incoming sensory data by virtue of the laminar organization of inter-area cortical connections. Namely, bottom-up driving connections target layer 4, located in the middle of the cortical column, and feedback connections target deep and superficial layers and avoid layer 4 [5-7]. Using lamina-resolved fMRI, we simultaneously measured the activity in three early visual cortical areas (V1-V3) that are recruited to represent stimulus information during visual working memory [8]. We observed item-specific working memory signals in early visual cortex. In V1, this item-specific activity was selectively present at deep and superficial cortical depths, avoiding the middle layers, and working-memory-related activity was present at all depths in V2 and V3. These results show for the first time the laminar organization of internally generated signals during visual working memory in the human visual system and provide new insights into how bottom-up and top-down signals in visual cortex are deployed.

Two-dimensional-NGC-SENSE-GRAPPA for fast, ghosting-robust reconstruction of in-plane and slice-accelerated blipped-CAIPI echo planar imaging.

PURPOSE: Ghosting-robust reconstruction of blipped-CAIPI echo planar imaging simultaneous multislice data with low computational load. METHODS: To date, Slice-GRAPPA, with "odd-even" kernels that improve ghosting performance, has been the framework of choice for such reconstructions due to its predecessor SENSE-GRAPPA being deemed unsuitable for blipped-CAIPI data. Modifications to SENSE-GRAPPA are used to restore CAIPI compatibility and to make it robust against ghosting. Two implementations are tested, one where slices and in-plane unaliasing are dealt in the same serial manner as in Slice-GRAPPA [referred to as one-dimensional (1D)-NGC-SENSE-GRAPPA, where NGC stands for Nyquist Ghost Corrected] and one where both are unaliased in a single step (2D-NGC-SENSE-GRAPPA), which is analytically and experimentally shown to be computationally cheaper. RESULTS: The 1D-NGC-SENSE-GRAPPA and odd-even Slice-GRAPPA perform identically, whereas 2D-NGC-SENSE-GRAPPA shows reduced error propagation, less residual ghosting when reliable reference data were available. When the latter was not the case, error propagation was increased. CONCLUSION: Unlike Slice-GRAPPA, SENSE-GRAPPA operates fully within the GRAPPA framework, for which improved reconstructions (e.g., iterative, nonlinear) have been developed over the past decade. It could, therefore, bring benefit to the reconstruction of SMS data as an attractive alternative to Slice-GRAPPA. Magn Reson Med 77:998-1009, 2017. © 2016 The Authors Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Resting-state subcortical functional connectivity in HIV-infected patients on long-term cART.

Despite long-term successful treatment with cART, impairments in cognitive functioning are still being reported in HIV-infected patients. Since changes in cognitive function may be preceded by subtle changes in brain function, neuroimaging techniques, such as resting-state functional magnetic resonance imaging (rs-fMRI) have become useful tools in assessing HIV-associated abnormalities in the brain. The purpose of the current study was to examine the extent to which HIV infection in virologically suppressed patients is associated with disruptions in subcortical regions of the brain in comparison to a matched HIV-negative control group. The sample consisted of 72 patients and 39 controls included between January 2012 and January 2014. Resting state functional connectivity was determined between fourteen regions-of-interest (ROI): the left and right nucleus accumbens, amygdala, caudate nucleus, hippocampus, putamen, pallidum and thalamus. A Bayesian method was used to estimate resting-state functional connectivity, quantified in terms of partial correlations. Both groups showed the strongest partial correlations between the left and right caudate nucleus and the left and right thalamus. However, no differences between the HIV patients and controls were found between the posterior expected network densities (control network density = 0.26, SD = 0.05, patient network density = 0.26, SD = 0.04, p = 0.58). The results of the current study show that HIV does not affect subcortical connectivity in virologically controlled patients who are otherwise healthy.

Cognitive Decline in Relation to Psychological Wellbeing and HIV Disease- and Treatment Characteristics in HIV-Infected Patients on cART: A One-Year Follow-Up Study.

The objectives of the current study were to examine cognitive decline in relation to psychological wellbeing, HIV disease and treatment characteristics and baseline variables over a one-year period of time in a group of HIV-infected patients on long term cART with undetectable viral load in comparison to a HIV-negative control group. Eighty-two of 95 patients and 43 of 55 controls who completed a baseline assessment for the Art-NeCo study underwent a follow-up neuropsychological assessment. A repeated-measure general linear model analysis was performed to compare the performance at follow-up in comparison to baseline between the patients and controls. Reliable change indices were computed as a measure of significant change in cognitive function. Compared to controls, patients overall performed worse on the domain speed of information processing. In the patient group a worse performance at follow-up was present for the verbal fluency domain compared to the controls, in the absence of a baseline group difference. For the executive function domain, no group differences were found at follow-up, but the patients performed worse than the controls at baseline. We found that cognitive decline was related to more frequent use of recreational drugs and a somewhat heightened level of irritability and more somatic complaints at baseline. However, the decliners did not differ from the non-decliners on any of the HIV-related variables.

High-resolution diffusion MRI at 7T using a three-dimensional multi-slab acquisition.

High-resolution diffusion MRI can provide the ability to resolve small brain structures, enabling investigations of detailed white matter architecture. A major challenge for in vivo high-resolution diffusion MRI is the low signal-to-noise ratio. In this work, we combine two highly compatible methods, ultra-high field and three-dimensional multi-slab acquisition to improve the SNR of high-resolution diffusion MRI. As each kz plane is encoded using a single-shot echo planar readout, scan speeds of the proposed technique are similar to the commonly used two-dimensional diffusion MRI. In-plane parallel acceleration is applied to reduce image distortions. To reduce the sensitivity of auto-calibration signal data to subject motion and respiration, several new adaptions of the fast low angle excitation echo-planar technique (FLEET) that are suitable for 3D multi-slab echo planar imaging are proposed and evaluated. A modified reconstruction scheme is proposed for auto-calibration with the most robust method, Slice-FLEET acquisition, to make it compatible with navigator correction of motion induced phase errors. Slab boundary artefacts are corrected using the nonlinear slab profile encoding method recently proposed by our group. In vivo results demonstrate that using 7T and three-dimensional multi-slab acquisition with improved auto-calibration signal acquisition and nonlinear slab boundary artefacts correction, high-quality diffusion MRI data with ~1mm isotropic resolution can be achieved.

The relationship between oscillatory EEG activity and the laminar-specific BOLD signal.

Electrophysiological recordings in animals have indicated that visual cortex γ-band oscillatory activity is predominantly observed in superficial cortical layers, whereas α- and ß-band activity is stronger in deep layers. These rhythms, as well as the different cortical layers, have also been closely related to feedforward and feedback streams of information. Recently, it has become possible to measure laminar activity in humans with high-resolution functional MRI (fMRI). In this study, we investigated whether these different frequency bands show a differential relation with the laminar-resolved blood-oxygen level-dependent (BOLD) signal by combining data from simultaneously recorded EEG and fMRI from the early visual cortex. Our visual attention paradigm allowed us to investigate how variations in strength over trials and variations in the attention effect over subjects relate to each other in both modalities. We demonstrate that γ-band EEG power correlates positively with the superficial layers' BOLD signal and that ß-power is negatively correlated to deep layer BOLD and α-power to both deep and superficial layer BOLD. These results provide a neurophysiological basis for human laminar fMRI and link human EEG and high-resolution fMRI to systems-level neuroscience in animals.

A hemodynamic model for layered BOLD signals.

High-resolution blood oxygen level dependent (BOLD) functional magnetic resonance imaging (fMRI) at the sub-millimeter scale has become feasible with recent advances in MR technology. In principle, this would enable the study of layered cortical circuits, one of the fundaments of cortical computation. However, the spatial layout of cortical blood supply may become an important confound at such high resolution. In particular, venous blood draining back to the cortical surface perpendicularly to the layered structure is expected to influence the measured responses in different layers. Here, we present an extension of a hemodynamic model commonly used for analyzing fMRI data (in dynamic causal models or biophysical network models) that accounts for such blood draining effects by coupling local hemodynamics across layers. We illustrate the properties of the model and its inversion by a series of simulations and show that it successfully captures layered fMRI data obtained during a simple visual experiment. We conclude that for future studies of the dynamics of layered neuronal circuits with high-resolution fMRI, it will be pivotal to include effects of blood draining, particularly when trying to infer on the layer-specific connections in cortex--a theme of key relevance for brain disorders like schizophrenia and for theories of brain function such as predictive coding.

Reducing slab boundary artifacts in three-dimensional multislab diffusion MRI using nonlinear inversion for slab profile encoding (NPEN).

PURPOSE: To propose a method to reduce the slab boundary artifacts in three-dimensional multislab diffusion MRI. METHODS: Bloch simulation is used to investigate the effects of multiple factors on slab boundary artifacts, including characterization of residual errors on diffusion quantification. A nonlinear inversion method is proposed to simultaneously estimate the slab profile and the underlying (corrected) image. RESULTS: Correction results of numerical phantom and in vivo data demonstrate that the method can effectively remove slab boundary artifacts for diffusion data. Notably, the nonlinear inversion is also successful at short TR, a regimen where previously proposed methods (slab profile encoding and weighted average) retain residual artifacts in both diffusion-weighted images and diffusion metrics (mean diffusion coefficient and fractional anisotropy). CONCLUSION: The nonlinear inversion for removing slab boundary artifacts provides improvements over existing methods, particularly at the short TRs required to maximize SNR efficiency. Magn Reson Med 76:1183-1195, 2016. © 2015 The Authors. Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Large dynamic range relative B1+ mapping.

PURPOSE: Parallel transmission (PTx) requires knowledge of the B1+ produced by each element. However, B1+ mapping can be challenging when transmit fields exhibit large dynamic range. This study presents a method to produce high quality relative B1+ maps when this is the case. THEORY AND METHODS: The proposed technique involves the acquisition of spoiled gradient echo (SPGR) images at multiple radiofrequency drive levels for each transmitter. The images are combined using knowledge of the SPGR signal equation using maximum likelihood estimation, yielding an image for each channel whose signal is proportional to the B1+ field strength. Relative B1+ maps are then obtained by taking image ratios. The method was tested using numerical simulations, phantom imaging, and through in vivo experiments. RESULTS: The numerical simulations demonstrated that the proposed method can reconstruct relative transmit sensitivities over a wide range of B1+ amplitudes and at several SNR levels. The method was validated at 3 Tesla (T) by comparing it with an alternative B1+ mapping method, and demonstrated in vivo at 7T. CONCLUSION: Relative B1+ mapping in the presence of large dynamic range has been demonstrated through numerical simulations, phantom imaging at 3T and experimentally at 7T. The method will enable PTx to be applied in challenging imaging scenarios at ultrahigh field. Magn Reson Med 76:490-499, 2016. © 2015 The Authors. Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Simultaneous multislice (SMS) imaging techniques.

Simultaneous multislice imaging (SMS) using parallel image reconstruction has rapidly advanced to become a major imaging technique. The primary benefit is an acceleration in data acquisition that is equal to the number of simultaneously excited slices. Unlike in-plane parallel imaging this can have only a marginal intrinsic signal-to-noise ratio penalty, and the full acceleration is attainable at fixed echo time, as is required for many echo planar imaging applications. Furthermore, for some implementations SMS techniques can reduce radiofrequency (RF) power deposition. In this review the current state of the art of SMS imaging is presented. In the Introduction, a historical overview is given of the history of SMS excitation in MRI. The following section on RF pulses gives both the theoretical background and practical application. The section on encoding and reconstruction shows how the collapsed multislice images can be disentangled by means of the transmitter pulse phase, gradient pulses, and most importantly using multichannel receiver coils. The relationship between classic parallel imaging techniques and SMS reconstruction methods is explored. The subsequent section describes the practical implementation, including the acquisition of reference data, and slice cross-talk. Published applications of SMS imaging are then reviewed, and the article concludes with an outlook and perspective of SMS imaging.